So, what does machine learning do and how can we use it? Simply speaking, it is very similar to skills the world’s best chess players have. You can anticipate every possible move and then program those into a computer, or a system, for it to implement it. If we look a little bit closer, we will even be able to recognise more layers piled on the top of each other. First and the biggest one is Artificial Intelligence (if, then, else) with lots of early excitement about “thinking machines”. More detailed one is Machine Learning (a thing labeller) where expert systems come of age. However, the most precise layer is Deep Learning (artificial neural networks). This one is considered as a proper “Cambrian Explosion” of AI.

What is more important though are five main practical uses for AI in market research. The first one is analysing language. Not that long ago running customer satisfaction trackers (CX analytics, social listening, survey coding), taking verbatim responses and figure out what customers are actually talking about was a common practise. Right now it is much easier to use platforms, which do the work for us and help us to find the underlying cause of complaints, reviews etc.

Second use lays within conversational interfaces. It applies to systems like chat and voice “bots”, natural language queries (i.e. understanding intent), intelligent assistants (i.e. relevance scoring or cross-document synthesis).

Number three is generating language. The most popular way to use it is when we create content (everything between data-driven text to automatic copy generation), automated reporting, intelligent moderation (shortcuts, reply suggestions).

Fourth practical use is visual analysis. This is particularly useful in decoding image content and context (brand use and consumer behaviour, speed up analysis time etc.), analysing video content and context and in visual salience models (there is no need for fresh data, it applies prior learning to new visual stimulus and provides massive cost reduction).

At least but not last, fifth use is the emotion AI. It helps researchers to recognise emotion in spoken language; in body language (it is brilliant especially in facial expression analysis and can be combined with other indicators of engagement) and emotion in skin, (it measures galvanic skin response via remote panel of sensor-equipped respondents).

There are good news for us humans too though. Obviously, AI is only as good as we are and there are things, which will never work without a human touch. For example, AI is terrible at empathy, context, narrative etc. and issues, which require human factors, are ethnography, customer closeness and user, experience reach.

Do not get too excited though! Predictions are that AI will be massively transformative for market research as if for any other industry or sector, mostly by reducing the cost per interview and increasing survey volumes at the same time. What this means is that more research will be done outside of research teams and agencies than within them. Overall advice to all suppliers is (again): recognise it is 2019 already and start investing into your tools, technologies, products etc. Otherwise, you will not only be stuck in the same place while your competitors are speeding ahead but in the eyes of your clients, you will actually be going backwards. That is never good news.

The above thoughts are taken from presentations by:

• Colin Strong, Head of Behavioural Science at Ipsos
• Panel Discussion: The Truth about Transforming Insight Teams - Ideal Plan vs Reality
• Mike Stevens, Founder at What Next Strategy & Planning